Glossary
Agentic commerce
Commerce where AI agents do part of the buying, the selling or the operating, with people in control.
What does agentic commerce mean?
Agentic commerce is commerce in which AI agents carry out tasks for a person or a business: finding and comparing products, placing orders, or running parts of operations such as answering stock questions and preparing documents. The agent acts through software interfaces, within permissions a person sets, and people approve the decisions that matter.
How agentic commerce works
On this page
The term covers two sides of the same shift.
- The buying side. A shopper, or a purchasing team, asks an AI assistant to find a product, compare options and complete the order. The agent reads catalogs, prices and stock, and checks out through an interface built for software.
- The operating side. A business lets agents work in its own systems: answer questions from live orders and inventory, flag exceptions, draft documents and reports, and hand decisions to a person.
What an agent needs
- A system of record it can read. Orders, stock, prices and customers in one place, current and consistent.
- An interface made for software. An API, or a tool protocol such as the Model Context Protocol (MCP), with named tools that return structured data.
- Permissions. The agent signs in as someone and gets that access and no more.
- Approvals. Reads can be open. Actions that spend money or move stock need a person or a rule to approve them.
- An audit trail. Every read and every change is recorded with who, or what, did it.
An example
An operations lead asks an assistant: "Which orders are late today, and why?" The agent calls a tool that lists unshipped orders, finds 12 past their ship date, sees that 9 of them wait on one SKU that is out of stock, and answers with the list and the cause. A person decides whether to split the orders or wait for the inbound purchase order. The numbers are illustrative.
Agentic commerce vs related terms
| Term | What it means |
|---|---|
| Conversational commerce | Selling through chat. A person still makes each choice |
| Automation and rules | Fixed if-this-then-that logic. It does not interpret a request |
| AI assistant or copilot | Answers and drafts when asked, inside one product |
| Agentic commerce | Agents plan steps, call tools across systems and act within set limits |
Common mistakes
- Giving an agent broad write access on day one. Start with reads, then add actions one at a time, with approvals.
- Letting agents read stale data. An agent that reads a nightly export gives an answer from yesterday with full confidence.
- No record of what the agent did. Without an audit trail, nobody can check or undo its work.
- Treating a chat window as the whole project. The work is in the data, the permissions and the tools behind it.
What agents can and cannot do in Kentro
Kentro is built to be used three ways: by people in the app, by Kentro AI, and by your own agents. Over the Kentro MCP server, an assistant such as Claude or Codex signs in as a user and reads orders, fulfillment, inventory, reports and reconciliation exceptions with the permissions of that user.
What agents cannot do: no MCP tool creates orders, buys labels, moves stock, creates purchase orders or changes settings. Software that needs to write uses the Channel API with a scoped API key, which writes core documents such as orders and shipments.
- Read tools over MCP, such as getFulfillmentPipeline and getInventorySummary
- The agent signs in as a user and inherits that role
- Writes go through the Channel API with a scoped key, not through MCP

Questions about agentic commerce
Is agentic commerce the same as an AI shopping assistant?
An AI shopping assistant is one form of it: an agent that finds, compares and sometimes buys products for a shopper. Agentic commerce is wider. It also covers agents that work for the seller, such as answering stock and order questions, flagging exceptions and preparing purchase documents. What they share is that software plans steps and calls tools, with a person setting the limits.
What is the Model Context Protocol (MCP)?
The Model Context Protocol is an open standard for connecting AI assistants to outside systems. A system publishes an MCP server with named tools, and an assistant that supports MCP can call those tools to read data or take actions, with the access its user was granted. It lets one connection work across many assistants, so a business does not build a separate integration for each one.
What should a business have in place before using AI agents in operations?
Three things. One current source of truth for orders, inventory and customers, because an agent repeats whatever the data says. Roles and permissions that already work for people, since agents should inherit them. And a clear line between what an agent may read, what it may propose and what it may do alone. Start with questions and reports before any action that spends money.
Can AI agents run a commerce business without people?
Not responsibly, today. Agents are good at reading large amounts of data, spotting exceptions and preparing work. They still make mistakes, and they act on whatever access they are given. The practical pattern is shared work: agents read, summarize and propose, and people approve anything that moves money, stock or customer commitments. The share that agents handle grows as controls and track records improve.
See what agents can do in Kentro
A dated list of every workflow: in the app, with Kentro AI, over MCP and over the API.

